bcachefs: Rework calling convention for marking overwrites

7 files changed, 95 insertions, 84 deletions

diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index 43dc2f270dc6..4cf728cea393 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -232,7 +232,7 @@ int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 	bch2_trans_init(&trans, c, 0, 0);
 
 	for_each_btree_key(&trans, iter, BTREE_ID_ALLOC, POS_MIN, 0, k, ret)
-		bch2_mark_key(c, k, 0, NULL, 0,
+		bch2_mark_key(c, k, 0, 0, NULL, 0,
 			      BCH_BUCKET_MARK_ALLOC_READ|
 			      BCH_BUCKET_MARK_NOATOMIC);
 
@@ -244,7 +244,8 @@ int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 
 	for_each_journal_key(*journal_keys, j)
 		if (j->btree_id == BTREE_ID_ALLOC)
-			bch2_mark_key(c, bkey_i_to_s_c(j->k), 0, NULL, 0,
+			bch2_mark_key(c, bkey_i_to_s_c(j->k),
+				      0, 0, NULL, 0,
 				      BCH_BUCKET_MARK_ALLOC_READ|
 				      BCH_BUCKET_MARK_NOATOMIC);
 
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index a458cfe0e92d..e43d48b8a342 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -171,7 +171,7 @@ static int bch2_gc_mark_key(struct bch_fs *c, struct bkey_s_c k,
 		*max_stale = max(*max_stale, ptr_stale(ca, ptr));
 	}
 
-	bch2_mark_key(c, k, k.k->size, NULL, 0, flags);
+	bch2_mark_key(c, k, 0, k.k->size, NULL, 0, flags);
 fsck_err:
 	return ret;
 }
@@ -418,7 +418,8 @@ static void bch2_mark_pending_btree_node_frees(struct bch_fs *c)
 
 	for_each_pending_btree_node_free(c, as, d)
 		if (d->index_update_done)
-			bch2_mark_key(c, bkey_i_to_s_c(&d->key), 0, NULL, 0,
+			bch2_mark_key(c, bkey_i_to_s_c(&d->key),
+				      0, 0, NULL, 0,
 				      BCH_BUCKET_MARK_GC);
 
 	mutex_unlock(&c->btree_interior_update_lock);
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index 9294137719df..6813eddd26f5 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -194,7 +194,7 @@ found:
 		       : gc_pos_btree_root(as->btree_id)) >= 0 &&
 	    gc_pos_cmp(c->gc_pos, gc_phase(GC_PHASE_PENDING_DELETE)) < 0)
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&d->key),
-			      0, NULL, 0,
+			      0, 0, NULL, 0,
 			      BCH_BUCKET_MARK_OVERWRITE|
 			      BCH_BUCKET_MARK_GC);
 }
@@ -266,11 +266,12 @@ static void bch2_btree_node_free_ondisk(struct bch_fs *c,
 {
 	BUG_ON(!pending->index_update_done);
 
-	bch2_mark_key(c, bkey_i_to_s_c(&pending->key), 0, NULL, 0,
-		      BCH_BUCKET_MARK_OVERWRITE);
+	bch2_mark_key(c, bkey_i_to_s_c(&pending->key),
+		      0, 0, NULL, 0, BCH_BUCKET_MARK_OVERWRITE);
 
 	if (gc_visited(c, gc_phase(GC_PHASE_PENDING_DELETE)))
-		bch2_mark_key(c, bkey_i_to_s_c(&pending->key), 0, NULL, 0,
+		bch2_mark_key(c, bkey_i_to_s_c(&pending->key),
+			      0, 0, NULL, 0,
 			      BCH_BUCKET_MARK_OVERWRITE|
 			      BCH_BUCKET_MARK_GC);
 }
@@ -1077,11 +1078,11 @@ static void bch2_btree_set_root_inmem(struct btree_update *as, struct btree *b)
 	fs_usage = bch2_fs_usage_scratch_get(c);
 
 	bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key),
-		      0, fs_usage, 0,
+		      0, 0, fs_usage, 0,
 		      BCH_BUCKET_MARK_INSERT);
 	if (gc_visited(c, gc_pos_btree_root(b->btree_id)))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key),
-				     0, NULL, 0,
+				     0, 0, NULL, 0,
 				     BCH_BUCKET_MARK_INSERT|
 				     BCH_BUCKET_MARK_GC);
 
@@ -1175,12 +1176,12 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as, struct btree *b
 	fs_usage = bch2_fs_usage_scratch_get(c);
 
 	bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
-			     0, fs_usage, 0,
+			     0, 0, fs_usage, 0,
 			     BCH_BUCKET_MARK_INSERT);
 
 	if (gc_visited(c, gc_pos_btree_node(b)))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(insert),
-				     0, NULL, 0,
+				     0, 0, NULL, 0,
 				     BCH_BUCKET_MARK_INSERT|
 				     BCH_BUCKET_MARK_GC);
 
@@ -2003,11 +2004,11 @@ static void __bch2_btree_node_update_key(struct bch_fs *c,
 		fs_usage = bch2_fs_usage_scratch_get(c);
 
 		bch2_mark_key_locked(c, bkey_i_to_s_c(&new_key->k_i),
-			      0, fs_usage, 0,
+			      0, 0, fs_usage, 0,
 			      BCH_BUCKET_MARK_INSERT);
 		if (gc_visited(c, gc_pos_btree_root(b->btree_id)))
 			bch2_mark_key_locked(c, bkey_i_to_s_c(&new_key->k_i),
-					     0, NULL, 0,
+					     0, 0, NULL, 0,
 					     BCH_BUCKET_MARK_INSERT||
 					     BCH_BUCKET_MARK_GC);
 
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 605be7badabb..eaa7005b35a8 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -810,23 +810,24 @@ void bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
 }
 
 static s64 ptr_disk_sectors_delta(struct extent_ptr_decoded p,
-				  s64 delta)
+				  unsigned offset, s64 delta,
+				  unsigned flags)
 {
-	if (delta > 0) {
-		/*
-		 * marking a new extent, which _will have size_ @delta
-		 *
-		 * in the bch2_mark_update -> BCH_EXTENT_OVERLAP_MIDDLE
-		 * case, we haven't actually created the key we'll be inserting
-		 * yet (for the split) - so we don't want to be using
-		 * k->size/crc.live_size here:
-		 */
-		return __ptr_disk_sectors(p, delta);
+	if (flags & BCH_BUCKET_MARK_OVERWRITE_SPLIT) {
+		BUG_ON(offset + -delta > p.crc.live_size);
+
+		return -((s64) ptr_disk_sectors(p)) +
+			__ptr_disk_sectors(p, offset) +
+			__ptr_disk_sectors(p, p.crc.live_size -
+					   offset + delta);
+	} else if (flags & BCH_BUCKET_MARK_OVERWRITE) {
+		BUG_ON(offset + -delta > p.crc.live_size);
+
+		return -((s64) ptr_disk_sectors(p)) +
+			__ptr_disk_sectors(p, p.crc.live_size +
+					   delta);
 	} else {
-		BUG_ON(-delta > p.crc.live_size);
-
-		return (s64) __ptr_disk_sectors(p, p.crc.live_size + delta) -
-			(s64) ptr_disk_sectors(p);
+		return ptr_disk_sectors(p);
 	}
 }
 
@@ -1005,7 +1006,8 @@ static int bch2_mark_stripe_ptr(struct bch_fs *c,
 }
 
 static int bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
-			    s64 sectors, enum bch_data_type data_type,
+			    unsigned offset, s64 sectors,
+			    enum bch_data_type data_type,
 			    struct bch_fs_usage *fs_usage,
 			    unsigned journal_seq, unsigned flags)
 {
@@ -1026,7 +1028,7 @@ static int bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k,
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		s64 disk_sectors = data_type == BCH_DATA_BTREE
 			? sectors
-			: ptr_disk_sectors_delta(p, sectors);
+			: ptr_disk_sectors_delta(p, offset, sectors, flags);
 		bool stale = bch2_mark_pointer(c, p, disk_sectors, data_type,
 					fs_usage, journal_seq, flags);
 
@@ -1115,7 +1117,8 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k,
 }
 
 int bch2_mark_key_locked(struct bch_fs *c,
-		   struct bkey_s_c k, s64 sectors,
+		   struct bkey_s_c k,
+		   unsigned offset, s64 sectors,
 		   struct bch_fs_usage *fs_usage,
 		   u64 journal_seq, unsigned flags)
 {
@@ -1136,11 +1139,11 @@ int bch2_mark_key_locked(struct bch_fs *c,
 			?  c->opts.btree_node_size
 			: -c->opts.btree_node_size;
 
-		ret = bch2_mark_extent(c, k, sectors, BCH_DATA_BTREE,
+		ret = bch2_mark_extent(c, k, offset, sectors, BCH_DATA_BTREE,
 				fs_usage, journal_seq, flags);
 		break;
 	case KEY_TYPE_extent:
-		ret = bch2_mark_extent(c, k, sectors, BCH_DATA_USER,
+		ret = bch2_mark_extent(c, k, offset, sectors, BCH_DATA_USER,
 				fs_usage, journal_seq, flags);
 		break;
 	case KEY_TYPE_stripe:
@@ -1171,14 +1174,14 @@ int bch2_mark_key_locked(struct bch_fs *c,
 }
 
 int bch2_mark_key(struct bch_fs *c, struct bkey_s_c k,
-		  s64 sectors,
+		  unsigned offset, s64 sectors,
 		  struct bch_fs_usage *fs_usage,
 		  u64 journal_seq, unsigned flags)
 {
 	int ret;
 
 	percpu_down_read(&c->mark_lock);
-	ret = bch2_mark_key_locked(c, k, sectors,
+	ret = bch2_mark_key_locked(c, k, offset, sectors,
 				   fs_usage, journal_seq, flags);
 	percpu_up_read(&c->mark_lock);
 
@@ -1194,8 +1197,11 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 {
 	struct bch_fs		*c = trans->c;
 	struct btree		*b = iter->l[0].b;
+	unsigned		offset = 0;
 	s64			sectors = 0;
 
+	flags |= BCH_BUCKET_MARK_OVERWRITE;
+
 	if (btree_node_is_extents(b)
 	    ? bkey_cmp(new->k.p, bkey_start_pos(old.k)) <= 0
 	    : bkey_cmp(new->k.p, old.k->p))
@@ -1204,35 +1210,33 @@ inline int bch2_mark_overwrite(struct btree_trans *trans,
 	if (btree_node_is_extents(b)) {
 		switch (bch2_extent_overlap(&new->k, old.k)) {
 		case BCH_EXTENT_OVERLAP_ALL:
+			offset = 0;
 			sectors = -((s64) old.k->size);
 			break;
 		case BCH_EXTENT_OVERLAP_BACK:
+			offset = bkey_start_offset(&new->k) -
+				bkey_start_offset(old.k);
 			sectors = bkey_start_offset(&new->k) -
 				old.k->p.offset;
 			break;
 		case BCH_EXTENT_OVERLAP_FRONT:
+			offset = 0;
 			sectors = bkey_start_offset(old.k) -
 				new->k.p.offset;
 			break;
 		case BCH_EXTENT_OVERLAP_MIDDLE:
-			sectors = old.k->p.offset - new->k.p.offset;
-			BUG_ON(sectors <= 0);
-
-			bch2_mark_key_locked(c, old, sectors,
-				fs_usage, trans->journal_res.seq,
-				BCH_BUCKET_MARK_INSERT|flags);
-
-			sectors = bkey_start_offset(&new->k) -
-				old.k->p.offset;
+			offset = bkey_start_offset(&new->k) -
+				bkey_start_offset(old.k);
+			sectors = -((s64) new->k.size);
+			flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
 			break;
 		}
 
 		BUG_ON(sectors >= 0);
 	}
 
-	return bch2_mark_key_locked(c, old, sectors, fs_usage,
-				    trans->journal_res.seq,
-				    BCH_BUCKET_MARK_OVERWRITE|flags) ?: 1;
+	return bch2_mark_key_locked(c, old, offset, sectors, fs_usage,
+				    trans->journal_res.seq, flags) ?: 1;
 }
 
 int bch2_mark_update(struct btree_trans *trans,
@@ -1250,10 +1254,12 @@ int bch2_mark_update(struct btree_trans *trans,
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return 0;
 
+	EBUG_ON(btree_node_is_extents(b) &&
+		!bch2_extent_is_atomic(insert->k, insert->iter));
+
 	if (!(trans->flags & BTREE_INSERT_NOMARK_INSERT))
 		bch2_mark_key_locked(c, bkey_i_to_s_c(insert->k),
-			bpos_min(insert->k->k.p, b->key.k.p).offset -
-			bkey_start_offset(&insert->k->k),
+			0, insert->k->k.size,
 			fs_usage, trans->journal_res.seq,
 			BCH_BUCKET_MARK_INSERT|flags);
 
@@ -1518,8 +1524,9 @@ out:
 }
 
 static int bch2_trans_mark_extent(struct btree_trans *trans,
-			struct bkey_s_c k,
-			s64 sectors, enum bch_data_type data_type)
+			struct bkey_s_c k, unsigned offset,
+			s64 sectors, unsigned flags,
+			enum bch_data_type data_type)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
@@ -1539,7 +1546,7 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 		s64 disk_sectors = data_type == BCH_DATA_BTREE
 			? sectors
-			: ptr_disk_sectors_delta(p, sectors);
+			: ptr_disk_sectors_delta(p, offset, sectors, flags);
 
 		ret = bch2_trans_mark_pointer(trans, p, disk_sectors,
 					      data_type);
@@ -1574,7 +1581,7 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
 }
 
 int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
-			s64 sectors, unsigned flags)
+			unsigned offset, s64 sectors, unsigned flags)
 {
 	struct replicas_delta_list *d;
 	struct bch_fs *c = trans->c;
@@ -1585,11 +1592,11 @@ int bch2_trans_mark_key(struct btree_trans *trans, struct bkey_s_c k,
 			?  c->opts.btree_node_size
 			: -c->opts.btree_node_size;
 
-		return bch2_trans_mark_extent(trans, k, sectors,
-					      BCH_DATA_BTREE);
+		return bch2_trans_mark_extent(trans, k, offset, sectors,
+					      flags, BCH_DATA_BTREE);
 	case KEY_TYPE_extent:
-		return bch2_trans_mark_extent(trans, k, sectors,
-					      BCH_DATA_USER);
+		return bch2_trans_mark_extent(trans, k, offset, sectors,
+					      flags, BCH_DATA_USER);
 	case KEY_TYPE_inode:
 		d = replicas_deltas_realloc(trans, 0);
 
@@ -1628,11 +1635,11 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return 0;
 
-	ret = bch2_trans_mark_key(trans,
-			bkey_i_to_s_c(insert),
-			bpos_min(insert->k.p, b->key.k.p).offset -
-			bkey_start_offset(&insert->k),
-			BCH_BUCKET_MARK_INSERT);
+	EBUG_ON(btree_node_is_extents(b) &&
+		!bch2_extent_is_atomic(insert, iter));
+
+	ret = bch2_trans_mark_key(trans, bkey_i_to_s_c(insert),
+			0, insert->k.size, BCH_BUCKET_MARK_INSERT);
 	if (ret)
 		return ret;
 
@@ -1640,7 +1647,9 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 						      KEY_TYPE_discard))) {
 		struct bkey		unpacked;
 		struct bkey_s_c		k;
+		unsigned		offset = 0;
 		s64			sectors = 0;
+		unsigned		flags = BCH_BUCKET_MARK_OVERWRITE;
 
 		k = bkey_disassemble(b, _k, &unpacked);
 
@@ -1652,35 +1661,32 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 		if (btree_node_is_extents(b)) {
 			switch (bch2_extent_overlap(&insert->k, k.k)) {
 			case BCH_EXTENT_OVERLAP_ALL:
+				offset = 0;
 				sectors = -((s64) k.k->size);
 				break;
 			case BCH_EXTENT_OVERLAP_BACK:
+				offset = bkey_start_offset(&insert->k) -
+					bkey_start_offset(k.k);
 				sectors = bkey_start_offset(&insert->k) -
 					k.k->p.offset;
 				break;
 			case BCH_EXTENT_OVERLAP_FRONT:
+				offset = 0;
 				sectors = bkey_start_offset(k.k) -
 					insert->k.p.offset;
 				break;
 			case BCH_EXTENT_OVERLAP_MIDDLE:
-				sectors = k.k->p.offset - insert->k.p.offset;
-				BUG_ON(sectors <= 0);
-
-				ret = bch2_trans_mark_key(trans, k, sectors,
-						BCH_BUCKET_MARK_INSERT);
-				if (ret)
-					return ret;
-
-				sectors = bkey_start_offset(&insert->k) -
-					k.k->p.offset;
+				offset = bkey_start_offset(&insert->k) -
+					bkey_start_offset(k.k);
+				sectors = -((s64) insert->k.size);
+				flags |= BCH_BUCKET_MARK_OVERWRITE_SPLIT;
 				break;
 			}
 
 			BUG_ON(sectors >= 0);
 		}
 
-		ret = bch2_trans_mark_key(trans, k, sectors,
-					  BCH_BUCKET_MARK_OVERWRITE);
+		ret = bch2_trans_mark_key(trans, k, offset, sectors, flags);
 		if (ret)
 			return ret;
 
diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h
index 5ab6f3d34137..799bfb3c96d8 100644
--- a/fs/bcachefs/buckets.h
+++ b/fs/bcachefs/buckets.h
@@ -251,14 +251,15 @@ void bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
 
 #define BCH_BUCKET_MARK_INSERT			(1 << 0)
 #define BCH_BUCKET_MARK_OVERWRITE		(1 << 1)
-#define BCH_BUCKET_MARK_BUCKET_INVALIDATE	(1 << 2)
-#define BCH_BUCKET_MARK_GC			(1 << 3)
-#define BCH_BUCKET_MARK_ALLOC_READ		(1 << 4)
-#define BCH_BUCKET_MARK_NOATOMIC		(1 << 5)
+#define BCH_BUCKET_MARK_OVERWRITE_SPLIT		(1 << 2)
+#define BCH_BUCKET_MARK_BUCKET_INVALIDATE	(1 << 3)
+#define BCH_BUCKET_MARK_GC			(1 << 4)
+#define BCH_BUCKET_MARK_ALLOC_READ		(1 << 5)
+#define BCH_BUCKET_MARK_NOATOMIC		(1 << 6)
 
-int bch2_mark_key_locked(struct bch_fs *, struct bkey_s_c, s64,
+int bch2_mark_key_locked(struct bch_fs *, struct bkey_s_c, unsigned, s64,
 			 struct bch_fs_usage *, u64, unsigned);
-int bch2_mark_key(struct bch_fs *, struct bkey_s_c, s64,
+int bch2_mark_key(struct bch_fs *, struct bkey_s_c, unsigned, s64,
 		  struct bch_fs_usage *, u64, unsigned);
 int bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *,
 			struct disk_reservation *, unsigned);
@@ -272,7 +273,8 @@ int bch2_mark_update(struct btree_trans *, struct btree_insert_entry *,
 void bch2_replicas_delta_list_apply(struct bch_fs *,
 				    struct bch_fs_usage *,
 				    struct replicas_delta_list *);
-int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c, s64, unsigned);
+int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c,
+			unsigned, s64, unsigned);
 int bch2_trans_mark_update(struct btree_trans *,
 			   struct btree_iter *iter,
 			   struct bkey_i *insert);
diff --git a/fs/bcachefs/ec.c b/fs/bcachefs/ec.c
index 438346b12901..bdb18c2a0b5c 100644
--- a/fs/bcachefs/ec.c
+++ b/fs/bcachefs/ec.c
@@ -1312,7 +1312,7 @@ int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 			break;
 		}
 
-		bch2_mark_key(c, k, 0, NULL, 0,
+		bch2_mark_key(c, k, 0, 0, NULL, 0,
 			      BCH_BUCKET_MARK_ALLOC_READ|
 			      BCH_BUCKET_MARK_NOATOMIC);
 	}
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 92867b5c078f..a7fc3fe4284a 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -295,7 +295,7 @@ retry:
 
 	if (split_compressed) {
 		ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k),
-					  -((s64) k->k.size),
+					  0, -((s64) k->k.size),
 					  BCH_BUCKET_MARK_OVERWRITE) ?:
 		      bch2_trans_commit(&trans, &disk_res, NULL,
 					BTREE_INSERT_ATOMIC|